ATAS

ATAS is an open-source agentic video generation framework that maintains a persistent world state across multi-shot films from plain-text screenplays.

ATAS is a world-state-centric framework for agentic video generation. It takes a plain-text screenplay and compiles it into bounded shot requests with explicit scene, camera, time, and event details. Every shot queries static scene memory and dynamic event memory to render context that conditions the video generator. Newly generated clips are reconstructed, registered, and validated through film review and state checks before updating the world. It outputs final MP4 videos alongside per-shot manifests and reports, utilizing models such as Pi3X and SAM 3.

ATAS at a glance

Best for: Software engineer
Runs on: open source
Pricing: Free
Released: October 2026
Category: AI video tool
Website: github.com/ziplab/atas

Key features

Compiles screenplays into bounded shot requests with camera and time data
Stores static scene geometry in a shared frame for multi-angle revisits
Indexes dynamic observations by scene, event, and time
Validates updates through film review, registration, and state checks
Outputs final.mp4, manifest.json, and report.md files
Supports resuming runs without repeating prior generations
Integrates with Pi3X and SAM 3 backbones

Use it for

Generating multi-shot films from plain-text screenplays
Re-filming past screenplay moments from alternative camera angles
Maintaining consistent environments across repeated scene revisits

Alternatives to ATAS

video2ctx — Provides video context with timestamped citations and uses 95% fewer tokens than transcripts.
Wan — Provides native 30-second video generation and pixel-perfect consistency.
Video2World — It evaluates coding agents on geometry, dynamics, and functionality across 222 tasks.
LiveFrame — It provides verified runtime, control, and stability reports with digest-bound ledgers.
Memvid — Single-file architecture replaces complex RAG pipelines with portable storage.
Bernini — Combines a semantic planner and DiT renderer for video generation and editing tasks.
MAGI-1 — Generates synchronized audio and visuals with high-fidelity camera control.
OneStreamer — It jointly learns to record evidence and respond through proactive generation.
FreeVideo — Runs MiniMax H3 locally on consumer GPUs with as little as 8 GB of VRAM.
Vibe Vision Video — Builds editable video projects using reusable templates and reference videos.
LingBot World — Generates physically plausible dynamic scenes with action conditioning.
TwelveLabs — It indexes video content for natural language search without manual metadata tags.

All ATAS alternatives